Identifying failure mechanisms based on a population of scan diagnostic reports

Abstract

Systems and techniques for identifying failure mechanisms based on a population of scan diagnostic reports is described. Given a population of scan diagnostic reports, a mixed membership model can be used for computing a topic distribution for each portion of each scan diagnostic report and a feature distribution for each topic. The failure mechanisms can be identified based on the topic distributions for the portions of the scan diagnostic reports and the feature distributions for the topics.

What is claimed is: 
     
       1. In a yield analysis software tool in a computer, a method for identifying failure mechanisms based on a population of scan diagnostic reports, wherein the population of scan diagnostic reports are generated by testing a set of manufactured integrated circuits, the method comprising:
 the yield analysis software tool in the computer using a mixed membership model to compute a topic distribution for each portion of each scan diagnostic report and a feature distribution for each topic, wherein the mixed membership model comprises (1) a first set of latent random variables that represent topic distributions for portions of scan diagnostic reports, (2) a second set of latent random variables that represent feature distributions for topics, and (3) a third set of observable random variables that represent features in the population of scan diagnostic reports; and 
 the yield analysis software tool in the computer identifying failure mechanisms based on the computed topic distribution for each portion of each scan diagnostic report and the computed feature distribution for each topic, wherein the identified failure mechanisms can be used to improve integrated circuit manufacturing yield. 
 
     
     
       2. The method of  claim 1 , wherein the mixed membership model is based on a Latent Dirichlet Allocation (LDA) model. 
     
     
       3. The method of  claim 1 , wherein each failure mechanism is one of:
 an interconnect open defect, an interconnect short defect, a defect in a standard cell, or a defect in a via. 
 
     
     
       4. The method of  claim 1 , wherein said identifying failure mechanisms comprises:
 computing aggregate weights for features based on the topic distribution for each portion of each scan diagnostic report and the feature distribution for each topic; and 
 sorting features in decreasing order of their aggregate weights. 
 
     
     
       5. The method of  claim 1 , wherein the topic distribution and the feature distribution are computed using a Bayesian approach. 
     
     
       6. A non-transitory computer-readable storage medium storing instructions for a yield analysis software tool that, when executed by a processor, cause the processor to perform a method for identifying failure mechanisms based on a population of scan diagnostic reports, wherein the population of scan diagnostic reports are generated by testing a set of manufactured integrated circuits, the method comprising:
 using a mixed membership model to compute a topic distribution for each portion of each scan diagnostic report and a feature distribution for each topic, wherein the mixed membership model comprises (1) a first set of latent random variables that represent topic distributions for portions of scan diagnostic reports, (2) a second set of latent random variables that represent feature distributions for topics, and (3) a third set of observable random variables that represent features in the population of scan diagnostic reports; and 
 identifying failure mechanisms based on the computed topic distribution for each portion of each scan diagnostic report and the computed feature distribution for each topic, wherein the identified failure mechanisms can be used to improve integrated circuit manufacturing yield. 
 
     
     
       7. The non-transitory computer-readable storage medium of  claim 6 , wherein the mixed membership model is based on a Latent Dirichlet Allocation (LDA) model. 
     
     
       8. The non-transitory computer-readable storage medium of  claim 6 , wherein each failure mechanism is one of: an interconnect open defect, an interconnect short defect, a defect in a standard cell, or a defect in a via. 
     
     
       9. The non-transitory computer-readable storage medium of  claim 6 , wherein said identifying failure mechanisms comprises:
 computing aggregate weights for features based on the topic distribution for each portion of each scan diagnostic report and the feature distribution for each topic; and 
 sorting features in decreasing order of their aggregate weights. 
 
     
     
       10. The non-transitory computer-readable storage medium of  claim 6 , wherein the topic distribution and the feature distribution are computed using a Bayesian approach. 
     
     
       11. A yield analysis system, comprising:
 a processor; and 
 a non-transitory computer-readable storage medium storing instructions that, when executed by the processor, cause the processor to perform a method for identifying failure mechanisms based on a population of scan diagnostic reports, wherein the population of scan diagnostic reports are generated by testing a set of manufactured integrated circuits, the method comprising:
 using a mixed membership model to compute a topic distribution for each portion of each scan diagnostic report and a feature distribution for each topic, wherein the mixed membership model comprises (1) a first set of latent random variables that represent topic distributions for portions scan diagnostic reports, (2) a second set of latent random variables that represent feature distributions for topics, and (3) a third set of observable random variables that represent features in the population of scan diagnostic reports; and 
 identifying failure mechanisms based on the computed topic distribution for each portion of each scan diagnostic report and the computed feature distribution for each topic, wherein the identified failure mechanisms can be used to improve integrated circuit manufacturing yield. 
 
 
     
     
       12. The yield analysis system of  claim 10 , wherein the mixed membership model is based on a Latent Dirichlet Allocation (LDA) model. 
     
     
       13. The yield analysis system of  claim 10 , wherein each failure mechanism is one of: an interconnect open defect, an interconnect short defect, a defect in a standard cell, or a defect in a via. 
     
     
       14. The yield analysis system of  claim 10 , wherein said identifying failure mechanisms comprises:
 computing aggregate weights for features based on the topic distribution for each portion of each scan diagnostic report and the feature distribution for each topic; and 
 sorting features in decreasing order of their aggregate weights. 
 
     
     
       15. The yield analysis system of  claim 10 , wherein the topic distribution and the feature distribution are computed using a Bayesian approach.